About
I'm Thomas Tucker.
I'm Senior Director of Engineering at G2, in Austin, Texas. I run engineering organizations where AI agents are part of the delivery system and every initiative is a business bet with a measurable hypothesis.
The record
Results from teams I've run.
- Engineers actively using AI tools at G2, in seven months
- 0 → 92%
- Partner-sourced revenue in two years, while traffic fell 38%
- 3.1×
- Revenue per delivered lead
- +32%
- Bugs down, developer efficiency up, after replacing manual QA with AI-assisted, developer-owned testing
- −12% / +20%
25 → 13. Led a restructure across the US, Spain and India, and the separation through G2's acquisition, with delivery kept on track.
How I lead
Care personally. Challenge directly. Let agents take the toil.
The principles I run teams on, drawn from research I trust and adapted for AI-era delivery.
01
Judge the system by its at-bats
We don't judge engineers on whether a feature moved revenue; markets decide that. We judge whether the system gives the business enough at-bats: shipped experiments, each with a hypothesis, a metric and a result. Failed hypotheses are celebrated on the board, not hidden.
Grounded in Amy Edmondson, Right Kind of Wrong
02
Every engineer is an agent manager
Directing agents is a skill: scoping work, giving context, reviewing output, knowing when to take the wheel. I coach it explicitly, the same way I coach code review or estimation.
Grounded in Ethan Mollick, Co-Intelligence
03
Agents take toil; people keep judgment
Agents plan, build and test. People approve plans above a complexity threshold, review every PR, and own every merge.
Grounded in Daniel Pink, Drive
04
Small teams, wider scope
When agents absorb throughput, human teams get smaller and each person's scope gets wider. A stream-aligned pod pairs with an agent squad; a platform team owns the harness, the gateway and the shared context, and runs them as products.
Grounded in Team Topologies; Camille Fournier, Platform Engineering
05
Measure to remove friction, never to rank
Usage data is team-level by default and visible to the engineer first. It's for finding the failing test loop, the missing permission, the expensive default, and fixing it in the harness.
Grounded in Kim Scott, Radical Candor
06
Spend like an owner
AI spend goes through a gateway with budgets per team, routing that keeps routine work off the most expensive model, and a forecast to fiscal year-end. Cheaper defaults beat after-the-fact policing.
Grounded in Will Larson, The Engineering Executive's Primer
Operating rules for agents
Written down, so your security team can read them.
- 1A complexity gate decides when a person must approve the plan before agents execute.
- 2Every PR gets a human review against a written checklist.
- 3Agents never merge, deploy to production or change ticket ownership.
- 4Agents read the same repo conventions engineers do, and both keep them current.
- 5Business context comes from a shared knowledge base, so decisions are made once and reused.
Restructure with care
Say it plainly and early, make decisions once, and protect the people who stay from a second round of uncertainty.
Tie work to the business daily
A daily view connects engineering work to revenue, qualifications and customer calls, so engineers see why the work matters.
Quality is the developer's job
Developer-owned, AI-assisted testing raised efficiency 20% and cut bugs 12%. The trick is training and tooling, not a mandate.
Résumé
Experience
Senior Director of EngineeringG2 · Austin, TX · Feb 2026 – Present
Own the pay-per-lead marketplace platform: advisor and partner portals, backend services, the legacy monolith and the SoftwareAdvice web properties. Lead engineering managers and their teams.
AI strategy and agent platform
- Took the team from zero AI spend in January 2026 to 92% of engineers actively using AI tools by August, with spend tracked and budgeted per team.
- Built an agent harness for software delivery: phase-gated skills (triage, plan, review, execute, test, PR, deploy, close), specialist subagents, an isolated git worktree per ticket, and repo conventions agents read and follow. Engineers use it daily.
- Launched a hosted agent squad that takes tickets from Jira through planning, build, test and PR review, with intake automation so tagging a ticket routes it to the right agents.
- Gave agents shared business and system context through MCP: a searchable knowledge base of business rules and decisions, and a queryable architecture graph of services, data stores and event flows.
- Set the operating rules for agents: complexity gates that decide when a human approves a plan, mandatory review checklists, and agents never merge or change ticket ownership.
- Led AI search optimization (AEO and GEO) to protect visibility as browsing shifts toward AI answers.
Revenue and growth
- Grew partner-sourced revenue 3.1× in two years, from 6% to 26% of marketplace revenue, offsetting a roughly 38% decline in organic and paid traffic.
- Raised revenue per delivered lead 32% through pricing, lead-form and market-expansion work.
Platform delivery
- Led the platform's separation from its former parent's infrastructure (domains, branding, authentication) against a hard contract-end deadline during the acquisition.
- Migrated bidding off a legacy PHP monolith to a Go service platform and run the monolith decommission program.
- Replaced legacy commission calculation with a new service covering time tracking, approvals and payroll export.
- Cut the buyer appointment flow from 9 steps to 2.
- Moved softwareadvice.com to Vercel: Core Web Vitals went from red to green across a site of millions of pages.
- Introduced daily business reporting that ties engineering work to revenue, qualifications and advisor calls.
Senior Director of Software EngineeringGartner Digital Markets · Feb 2025 – Feb 2026
- Led a 25-person, three-region organization with four engineering managers (US, Spain, India, and a dedicated QA team).
- Restructured to 13 people and two managers in response to declining revenue, done with care and with delivery kept on track.
- Replaced the dedicated QA function and a large device-testing contract with developer-owned, AI-assisted test automation and training. Reinvested part of the savings in observability. Developer efficiency rose 20% and bugs fell 12%.
Director of Software EngineeringGartner Digital Markets · Apr 2022 – Jan 2025
- Led engineering for the pay-per-lead platform and SoftwareAdvice web applications as marketplace revenue grew about 15%.
- Developed engineering managers and standardized on Terraform, EKS and server-rendered Next.js, with Core Web Vitals measured in production.
Software Engineering Manager · Lead Software EngineerGartner Digital Markets · Jan 2019 – Apr 2022
- Promoted from lead engineer to manager of two Next.js teams while the platform's revenue grew more than 50%.
- Cut production bugs to near zero in a server-rendered Angular application through unit coverage and end-to-end testing.
Earlier experience
- Silvercar, Senior Software Engineer, 2017–2018. Rewrote the booking app in React and Redux in six months.
- Alert Logic, Principal Engineer, 2014–2017. Technical lead for a UX and UI engineering team.
- Dun & Bradstreet, Lead UI Developer, 2011–2013. Led eight UI developers.
- Earlier: Rapid7, Prism.fm, Pervasive Software, PROACE, Hover Studios (founder)
AI delivery
- Agent harnesses
- MCP servers
- Claude Code
- LiteLLM gateways
- LangSmith tracing and evals
- AI spend governance
Platform
- Next.js
- Go services
- Terraform
- AWS EKS
- Vercel
- Datadog and OpenTelemetry
Leadership
- Managing managers
- Distributed teams (US, Spain, India)
- Restructuring with care
- Acquisitions and separations
Education and credentials
- Texas Tech University: BA in English, Music and Political Science
- Patent: Electronic device protection system and method
Want this story on your stage?
I reply within two business days.